Google claims to have overcome a major bottleneck in quantum computing
Decrypt
1h ago
Ai Focus
Google Quantum AI indicates that a technology called "magic state cultivation" has reduced the error rate by about 40 times on its Willow chips, achieving a magic state fidelity of 99.99%. Google claims that this progress is expected to reduce the cost of manufacturing key resources in quantum computing, but there is still a significant gap before cracking encryptions such as RSA-2048.
Helpful
No.Help

Google says it has turned one of the most promising concepts in quantum computing into a technology that can actually run on chips.

Google Quantum’s AI team published a paper on Science on Thursday, stating that a technique called “magic state cultivation” magic state cultivation reduced the error rate by about 40 times on their Willow processors.

Making quantum computers more affordable is of great importance not only for physics laboratories, as it would bring forward the day when they can perform useful tasks such as designing drugs and new materials, but it would also bring us one step closer to potentially cracking the encryption that protects bank login information and Bitcoin wallets.

Quantum computers operate based on quantum bits ( qubit ) – quantum bits can exist in a superposition state of both 0 and 1 simultaneously, allowing the machine to explore many answers in parallel. However, they are also extremely fragile: a small amount of heat or electrical noise can flip a quantum bit and ruin the computation. Therefore, the entire industry is focused on error correction, which involves distributing information across multiple quantum bits in order to detect and correct errors.

Error correction can handle simple calculations well, but truly useful programs require a different type of more complex calculations, which cannot be protected in the same way. To perform these calculations, the machine must first create a "magical state"; they are like disposable fuel cells for these calculations: each magical state is consumed the moment it is used.

Google claims that producing a clean quantum state has always been extremely costly. According to Google, some estimates suggest that this process could consume up to 95% of the total capacity of a quantum computer, leaving very limited space for actual computations.

Physicists Sergei Bravyi and Alexei Kitaev proposed a standard solution as early as 2004 – Magic State Distillation ( magic state distillation ), which involves purifying many noisy magic states into a few clean magic states. This method is effective, but it requires massive quantum bit “factories”, which do nothing else but this task.

In 2024, Google researchers proposed a cheaper approach, which they named "magic state cultivation." Unlike factories, this technique is more akin to a filter: a rough "magic state" is injected into a very small area of the chip, and then a series of checks are performed. Each attempt that is found to be defective is discarded, leaving only the survivors that continue to "grow." The paper even suggests that further refinement "may never be necessary in practice."

Google stated: "On Google Quantum's AI Willow processor, our protocol achieved an astonishing 40-fold reduction in error compared to the untrained state. We also obtained the T-state, with a logical fidelity record of 99.99%."

Only about 8% of the attempts pass all checks, which means that the chip discards more than nine-tenths of the attempts in order to maintain a clean state.

Cultivating technology is one of the key techniques in Gidney's estimates for 2025. This estimate suggests that RSA-2048 encryption can be cracked using less than 1 million noisy quantum bits, which is a 20-fold reduction from the 20 million estimated in 2019. Prior to this experiment, cultivating technology had only been tested in theory and computer simulations.

In March, Google further tightened this estimate. At that time, a white paper co-authored by Google's quantum researcher AI, Ethereum Foundation researcher Justin Drake, and Stanford cryptographer Dan Boneh stated that the cost of cracking Bitcoin's elliptic curve cryptography was less than 500,000 physical quantum bits. Elliptic curve cryptography is a mathematical mechanism that prevents anyone from deriving private keys from public addresses. This result builds upon Google's experiment in October last year Quantum Echoes – which was the first verified quantum acceleration on real hardware.

However, this does not mean that someone is stealing wallets. In this experiment, at most 54 out of Willow 105 quantum bits were used, and the paper points out that to break down a RSA-2048 key, a magic state with an error of about one part in a trillion is required, which is approximately a million times cleaner than the level just achieved by Google.

Google stated that the next step is to scale up the development of this technology. The team expects that if the underlying hardware noise of Willow is reduced by half, the magic-state error will drop to about two in a hundred thousand, as the errors in this process decrease with the cube of the chip's noise level.

Tip
$0
Like
0
Save
0
Views 29
HKWDB reminds readers to view blockchain rationally, stay aware of risks, and beware of virtual token issuance and speculation. All content on this site represents market information or related viewpoints only and does not constitute any form of investment advice. If you find sensitive content, please click“Report”,and we will handle it promptly。
Submit
Comment 0
Hot
Latest
No comments yet. Be the first!
Related
Bitcoin, BNB, HYPE, and ADA Price Analysis: Can Buyers Regain Control?
Bitcoin tests support around $80,000, BNB hovers above $700, HYPE holds the key support around $77, while ADA slides towards the 100-day moving average. The article analyzes the recent trends of four crypto assets, their moving average positions, and possible support and resistance zones.
U.Today
·2026-10-16 08:16:35
5
OpenAI may struggle to meet advertising revenue targets and is aggressively recruiting talents from Google and Meta
OpenAI has intensively launched advertising products, lowered pricing, and undertaken large-scale recruitment in recent months, driving a rapid increase in the number of advertising clients. By the end of August, the annual advertising revenue reached $1 billion. However, reports citing informed sources and research institutions suggest that the company's goal of achieving $2.4 billion in advertising revenue by 2026, which was set earlier this year with investors, is likely to be difficult to achieve.
The Block
·2026-10-16 07:56:18
19
Top 20 US stock traders on October 16: CoreWeave rose 5.94% Jensen Huang says there's no need for NVIDIA to cover its losses
Among the top 20 largest U.S. stock trades, NVIDIA fell 1.84% with trading volume of $27.415 billion; Micron Technology rose 1.87% with trading volume of $35.017 billion; CoreWeave rose 5.94% with trading volume of $8.389 billion. Jensen Huang stated that CoreWeave's financial strength is increasing, and there is no need for NVIDIA to provide support for them.
The Block
·2026-10-16 07:46:40
18
The Dow Jones rose by over 550 points, medical and consumer stocks strengthened, chip stocks continued to decline, Chinese concept stocks generally rose, and U.S. oil prices increased by more than 3%
U.S. stock market indexes closed higher on Thursday, with the Dow Jones Industrial Average rising more than 1%. In September, the PPI in the U.S. exceeded expectations, but U.S. Treasury yields fell, and funds shifted to the healthcare and consumer sectors, while chip stocks continued to face pressure. The NASDAQ Golden Dragon China Index rose 1.96%, WTI crude oil rose more than 3%, and gold and silver rebounded.
The Block
·2026-10-16 07:46:37
16
Microsoft AI claims that a person using one-quarter of their computing power defeated OpenAI
Microsoft CEO Mustafa Suleiman stated that the company's internal model is superior to a similar product from OpenAI, and only a quarter of the computing power used by that product was required for training. On that day, Microsoft also announced MAI-Base-2 and claimed that it outperforms some OpenAI and DeepSeek models in private tests as well.
Businessinsider
·2026-10-16 07:46:36
18
View More